ContiMorph: An unsupervised learning framework for cardiac motion tracking with time-continuous diffeomorphism

Area of Science:

• Medical imaging analysis
• Biomedical engineering
• Machine learning in healthcare

Background:

• Cardiac motion tracking is vital for assessing heart function and diagnosing cardiovascular conditions.
• Current methods using scaling-and-squaring (SS) integration produce discrete motion fields, limiting temporal continuity and accuracy.
• There is a need for advanced techniques to fully leverage the temporal nature of cardiac motion for improved tracking.

Purpose of the Study:

• To introduce ContiMorph, an unsupervised learning framework for time-continuous cardiac motion tracking.
• To enhance the accuracy and temporal fidelity of cardiac motion analysis in image sequences.
• To provide a robust method for evaluating cardiac function and diagnosing diseases.

Main Methods:

• Developed ContiMorph, integrating a frame-aware U-Net with a time-embedded transformer for continuous intra-frame motion.
• Composed continuous intra-frame fields into time-continuous Lagrangian motion fields for precise tracking.
• Implemented a time-continuous Lagrangian motion constraint with semigroup regularization to ensure diffeomorphic topology and leverage temporal information, eliminating SS integration.

Main Results:

• ContiMorph demonstrated superior performance compared to existing methods on cardiac MRI and echocardiography datasets.
• Achieved state-of-the-art results across various imaging modalities, showcasing its versatility.
• The framework effectively learns temporally continuous motion fields and maintains topological consistency.

Conclusions:

• ContiMorph represents a significant advancement in time-continuous cardiac motion tracking.
• The unsupervised learning approach offers improved accuracy and robustness for cardiac image analysis.
• This method has the potential to enhance clinical diagnosis and patient management for cardiovascular diseases.